Page 60 - Proceedings of the 2018 ITU Kaleidoscope

2018 ITU Kaleidoscope Academic Conference

challenges to give future research directions. Finally, we application may have to identify object in real-time but also
summarize our work in Section 6. such delay could lead to disastrous consequences. Thus, it
is necessary to devise alternative solutions to the current
2. BACKGROUND store-and-process later systems such that processing and
intelligent decision-making based on such data can be done
One major enabler of AIMS is the integration of 5G and close to the data sources in real-time.
Cloud computing, enabling the 5G applications to leverage
the abundant compute and storage power of the geo- Additionally, various solutions and concepts have recently
distributed Cloud data centers. been proposed to address this problem, from federated
clouds, to Edge computing [8]. The federated clouds are a
One of the widely researched challenges relates to how collection of heterogeneous infrastructures that may span
heterogeneous services and their operating platforms can the entire globe and requires that data from the IoT devices
interoperate on the same network. However, various be transmitted to the cloud data centers. Thus, this
research enthusiasts are aggressively addressing these “5G architecture still depends on the Internet and public
Vertical” challenges to enable the development of network telecommunication infrastructure with very high latency
slicing, multi tenancy, network programmability [5]. One of and bandwidths requirements. Fog, on the other hand, aims
the main weaknesses of the solutions in this regard is that to provide a system level horizontal architecture that
they still rely on transporting humongous IoT data across distributes computing, storage, control and networking
the 5G networks to various cloud data centers for storage or functions closer to the users in the Thing-Cloud continuum
processing. We have identified the negative impacts and [10]. ROOF computing is closely related to Fog computing
challenges of this model as follows: in that it provides highly distributed pervasive and
virtualized platform data/processing to a central cloud data
• Latency sensitive nature of the Edge based application center. However, ROOF computing has been proposed to
services necessitates that real-time decisions based on
provide highly functional, secure and scalable IoT. It
the acquired data from the Edge devices requires
promises interoperable connectivity for variety of Things
mechanisms for real-time processing of data for real- under the ROOF, context information and decisions for
time intelligence [6]. How do we design, model and
taking actions in real-time, information management and
expose these intelligent services for decision making at efficient connectivity to the Cloud and Service as well as
the Edge of Things to address latency related problems
efficient network design [4], [11]. This reduces
of AI services across the 5G networks?
communication delays and the size of data that needs to be
• Intelligent decision making at the Edge of Things migrated across the 5G to the cloud data centers.
introduces new AI dimension to IoT services such as
real-time local processing of IoT data for quick 3. AI AS MICROSERVICES (AIMS) AT THE EDGE
intelligent decision making without necessarily OF THINGS
transporting the heavy data through the expensive 5G
networks. The challenge here is how do we develop To deploy data-driven intelligent capability at the 5G
data-centric IoT Services in which AI is a first-class networks, AI in various forms of machine learning
design element? algorithms, such as the deep learning, must be infused into
the Edge-Cloud platform components. Thus, the 5G
Indeed, to take advantage of interoperable IoT platforms capabilities should be equipped with tools that allow
over 5G networks, IoT applications should be driven by AI intelligent services to be composed as data-driven
deployed as autonomous microservices, essentially microservices [12], [13]. The rationale is to address the
implementing the DIKW (Data, Information, Knowledge weaknesses of the current monolithic Cloud based AI
and Wisdom) at the edge of IoT [7]. Additionally, as services, which cannot meet the requirements of real-time
interoperable IoT based platforms are being deployed and ultra-low delay sensitive 5G applications. Rather than
through various use cases such as Smart City applications, shipping the data to the cloud data centers where AI
Smart Manufacturing, etc., transporting huge volume of algorithms are applied to incorporate intelligent decision-
data from the IoT edge to the geo-distributed centralized making capabilities into 5G applications, these AI
Cloud data centers for processing is not only efficient in algorithms can be implemented and deployed closer to the
terms of communication bandwidth and energy sources of the IoT data and users by factoring the AI
consumption but also cannot support ultra-low latency functionality into smaller functions that can be
applications [8]. implemented as distributed microservices [14]. We
proposes a hierarchically integrated infrastructure spanning
the ROOF, Fog and Cloud computing platforms (Figure 1),
With these ultra-low-delay sensitive applications, the
current solutions are obviously not practicable. For example, to exploit resources at the Edge of Things (ROOF and Fog
Computing resources) and Cloud data centers, as well as
a security surveillance application requires real-time
processing of huge live video data, which is transmitted to microservice concepts to incorporate AI capabilities into
IoT applications. The microservice concept allows the
the Cloud data centers for processing before intelligent
decisions can be made [9]. This approach will not only be decomposition or factoring of the current monolithic AI
services (which are deployed only on the centralized Cloud
impossible to meet the latency requirements as such

– 44 –